Key words: Credit-to-GDP gap, countercyclical capital ...
Transcript of Key words: Credit-to-GDP gap, countercyclical capital ...
1
The credit-to-GDP gap and complementary indicators for macroprudential policy:
Evidence from the UK1
Julia Giese2, Henrik Andersen
3, Oliver Bush
4, Christian Castro
5, Marc Farag
6 and Sujit
Kapadia7
Abstract: The financial crisis has demonstrated the need for a set of macroprudential policy
tools that can be used to mitigate systemic risk. Focusing on the UK, our paper reviews the
performance of the Basel III credit-to-GDP gap which, alongside judgement, is to be used as
a reference guide in setting the countercyclical capital buffer. We find that this measure
worked well in providing an advance signal of past UK episodes of banking system distress.
But this does not guarantee future signalling success. We therefore evaluate some conceptual
shortcomings of the credit gap and suggest complementary indicators.
Key words: Credit-to-GDP gap, countercyclical capital buffer, early warning indicators, financial
crises, macroprudential policy
1 Introduction
The financial crisis has demonstrated the need for a broader set of policy tools that can be
used to mitigate systemic risk. Time-varying macroprudential policy aims to enhance the
resilience of the banking system and over-exuberance in the supply of credit by discouraging
the build-up of financial imbalances that might otherwise have led to a systemic banking
crisis (Bank of England (2009), CGFS (2010), Bank of England (2011b), Borio (2011),
FSB/IMF/BIS (2011) and IMF (2011) among others). The Basel III framework has proposed
the countercyclical capital buffer as one tool that would be applied at the aggregate level of
national banking systems. The countercyclical capital buffer is an extension to the regulatory
capital framework for banks which policymakers will be able to adjust in a time-varying way.
Other time-varying macroprudential tools, including sectoral capital requirements, leverage
ratios and liquidity requirements, will also likely be important.
To help guide the activation and release of the countercyclical capital buffer, the Basel
Committee on Banking Supervision has suggested that it should be raised when a country’s
credit-to-GDP ratio exceeds its long-run trend by two percentage points.8 This BCBS guide
will serve as a common international guideline for policymakers taking buffer decisions,
alongside other indicators and judgement.
1 The views expressed in this paper are those of the authors, and not necessarily those of the Bank of England or the Financial Policy
Committee. We are grateful to Timothy Richards for research assistance. We also thank participants and discussant Vladimir Borgy at the Conference on Macroeconomic and Financial Vulnerability Indicators in Advanced Economies in Strasbourg in September 2012, seminar
participants at the Bank of England, an anonymous referee, David Aikman, Oliver Burrows, Jas Ellis, Hanno Stremmel and Victoria Saporta
for helpful comments. 2 Bank of England, Threadneedle Street, London EC2R 8AH, [email protected], corresponding author 3 Norges Bank (seconded to the Bank of England while this was being prepared) 4 Bank of England, Threadneedle Street, London EC2R 8AH, [email protected]. 5 Banco de Espana (working for the Bank of England while this was being prepared) 6 Bank of England, Threadneedle Street, London EC2R 8AH, [email protected] 7 Bank of England, Threadneedle Street, London EC2R 8AH, [email protected] 8 See BCBS (2010).
2
This paper considers the credit-to-GDP gap and other indicators which may help to guide the
setting of macroprudential capital requirements at both an aggregate and sectoral level in a
UK context; indicators for other potential areas of macroprudential policy, such as liquidity
requirements or structural policies, are left for future work. The Financial Policy Committee
of the Bank of England has published lists of core indicators for the countercyclical capital
buffer and sectoral capital requirements (Bank of England (2013)). Our analysis should be
viewed as complementary: we discuss a subset of the indicators in Bank of England (2013) in
more detail with a focus on indicators for which we have long time series available. We take
the credit-to-GDP gap as a starting point, arguing that the gap as defined by the BCBS
worked well in providing an advance signal of past UK crises. This is despite us using a real-
time measure (see Edge and Meisenzahl (2011) for a contrasting discussion of this issue).
But, as noted in Bank of England (2013), no single indicator can ever provide a perfect guide
to systemic risks or the appropriate policy responses, given the complexity of financial
interlinkages, the tendency for the financial system to evolve over time and time lags before
risks become apparent. Policymakers will also need to monitor a wide and time-varying set
of indicators, depending on the emerging risks. In this paper, we therefore discuss additional
indicators that might complement the credit-to-GDP gap, both in the context of having been
useful in past crises in the UK and elsewhere and in helping to address some of the credit-to-
GDP gap’s conceptual shortcomings. In particular, we argue that the sources and quality of
credit are important, suggesting the need not only for an aggregate capital tool such as the
countercyclical capital buffer but also sectoral tools such as sectoral capital requirements.
Finally, we discuss indicators that might help determine when to release the capital buffers.
In discussing other indicators, we take guidance from the literature based on cross-country
analyses (for example, Kaminsky and Reinhart (1999), Drehmann et al. (2010), Drehmann et
al. (2011), Borio and Lowe (2002, 2004) and Barrell et al. (2010)). Cross-country studies
have the obvious advantage of relying on more observations which gives more rigour to
understanding the indicators that have tended to signal banking crises in the past. However,
they also have drawbacks. For example, there may be little gained by including countries
with different institutional arrangements and financial structures in the same panel analysis.
Moreover, data definitions may not be homogeneous across countries and time series
employed in panel approaches tend to be limited. By taking the lessons from the cross-
country literature but focusing on the UK only for our own analysis, we have the advantage
of being less limited by data availability, can focus more closely on individual series and are
able to provide greater context. But the number of crises is small compared with cross-
country studies: the UK experienced three episodes of banking system distress over the past
50 years. Our dataset begins in the late 1960s for most series, covering the secondary
banking crisis from Q4 1973 to Q4 1975, the small banks’ crisis from Q3 1990 to Q2 1994
and the current crisis from Q3 2007 onwards.9 Some common elements were evident in all
three episodes: rapid credit expansions fuelled property price booms and financial institutions
encountered liquidity problems as funding markets dried up.
9 For the secondary banking crisis see Reid (1982) and for the small banks’ crisis Logan (2000).
3
The secondary banking crisis was the result of so-called unregulated ‘fringe’ institutions
funding themselves in money markets and investing these into, largely, commercial property
developments. Tightening monetary policy in 1973 contributed to the drying up of short-
term funding in money markets and the emergence of the problems at fringe institutions. To
stop problems passing into the regulated banking system, a ‘lifeboat’ was set up whereby
regulated banks were given responsibility for supporting ailing fringe institutions, in
cooperation with the Bank of England. The small banks’ crisis was not dissimilar. Again,
asset prices, particularly in commercial and residential real estate, rose, alongside credit. As
monetary policy was tightened, property markets turned and smaller banks, faced with losses,
lost access to money markets. From mid-1991, the Bank of England provided liquidity
support to a few small banks as their failure might have caused further disruption in funding
markets. Rapid commercial and residential property price growth was also a feature ahead of
the current crisis. This was accompanied by a surge in lending within the financial system as,
for example, mortgages were securitised and sold on, increasing interconnectedness between
institutions. And banks’ funding once again proved a weak point. As global wholesale
funding markets dried up in 2007-8, banks experienced liquidity problems that soon turned
into solvency problems. But this time, problems were not limited to small banks – two of the
UK’s largest banks, LBG and RBS, had to seek government support.
2 The credit-to-GDP gap for the UK
Minsky (1972) and Kindleberger (1978) both argue that credit booms tend to sow the seeds
of crises. In terms of empirical underpinning, several studies, including Alessi and Detken
(2009), Borgy et al. (2009), Borio and Lowe (2002, 2004), Drehmann et al. (2010),
Drehmann et al. (2011) and Schularick and Taylor (2012), have found that indicators of
excess credit growth are powerful in providing advance signals of financial crises.
Dell’Ariccia et al. (2012) find that a third of credit booms are followed by crises and three
fifths are followed by a period of economic underperformance (measured by the difference
between GDP growth relative to its long-run trend) in the six years following the end of the
boom. And micro evidence in Mendoza and Terrones (2008) suggests that bank capital
adequacy standards tend to fall during credit booms (at least in EMEs).
But defining a credit boom is tricky. The BCBS has suggested deviations of a credit-to-GDP
ratio from its long-term trend for this purpose. The credit measure it recommends is broad
credit to the household and private non-financial corporate (PNFC) sector, including non-
banks and lending from abroad. The BCBS further proposes that the long-run trend be
calculated by a one-sided, or ‘real-time’, Hodrick-Prescott (HP) filter with a smoothing
parameter of 400,000.10
Chart 1 illustrates this measure for the UK, showing that the broad measure would have
signalled the need to tighten the countercyclical capital buffer ahead of the past three
10 An HP filter is used to separate slow-moving trends from cycles in the data. The trend is calculated by minimising a function containing
the deviation of the trend from the data and the variation in the growth rate of the trend. A smoothing parameter (the lambda) determines the weight put on the proximity of the trend to the data and the variation in the trend.
4
episodes of banking system distress in the UK. Although the data needed to calculate the
broad measure are only available since 1987 in the UK, we have constructed an
approximation of the credit gap back to the 1960s by extrapolating the credit series using data
on narrow credit (UK-resident bank lending to resident households and PNFCs) that go back
to 1963. We also show the narrow series given that the data included in the broad series on
lending to PNFCs is based on approximations and the narrower data are of higher quality for
the UK. Both series do well, although the narrow series would have suggested a more
gradual tightening of the buffer ahead of the current crisis.
Chart 1: Gap of broad and narrow HH and
PNFC credit-to-GDP relative to long-term
trend(a)(b)(c)
Chart 2: Initial and revised estimates of the
credit-to-GDP ratio(a)
Source: ONS, Bank of England and Bank calculations.
(a) The credit-to-GDP gap is calculated as the percentage point
difference between the credit to GDP ratio and its long-term trend, where the trend is based on a one-sided/two-sided HP filter with a
smoothing parameter of 400,000. (b) Broad credit is defined here as debt claims on the UK private non-financial sector. This includes all liabilities of the household
and not-for-profit sector and private non-financial corporations’
loans and debt securities excluding derivatives, direct investment loans and loans secured on dwellings. ONS data are not available
before 1990. Before then, stable relationships between the ONS
household and private non-financial corporation debt data and Bank of England household and private non-financial corporation lending
data are assumed and the ONS household and private non-financial
corporation debt series is assumed to grow at the same rate as the Bank of England household and private non-financial corporation
lending series.
(c) Narrow credit comprises UK-resident monetary financial institution s’ sterling and foreign currency lending to households
and PNFCs.
Source: ONS, Bank of England and Bank calculations. (a) Credit is defined here as debt claims on the UK private non-
financial sector. This includes all liabilities of the household sector
and private non-financial corporations’ loans and debt securities.
Next, we discuss three empirical challenges raised in relation to the credit-to-GDP gap and
how these relate to the usefulness of the credit-to-GDP gap in assessing risk to the UK
financial system.
Challenge i): The role of data revisions
Building on work by Orphanides and van Norden (2002), Edge and Meisenzahl (2011)
question the reliability of the credit-to-GDP gap in real time. They argue that revisions to the
underlying data used to calculate the credit-to-GDP ratio may lead to significant policy error,
-20
-10
0
10
20
30
1965 1971 1977 1983 1989 1995 2001 2007
Broad private non-financial credit-to-GDP gap
Narrow private non-financial credit-to-GDP gap
Per cent of GDP
150
160
170
180
190
200
210
220
230
240
2004 2005 2006 2007 2008 2009 2010 2011 2012
Initial ratio Latest ratioPer cent of
GDP
5
though they conclude that such revisions are not large in the case of the US and are therefore
unlikely to lead to mistakes. But this may be a more significant problem in the UK. To
check this, we analyse real-time data on GDP and broad credit to households and PNFCs.
Our dataset includes (most) vintages from December 2004 onwards. As Chart 2 shows,
revisions over this period were significant for the credit-to-GDP ratio. But it does not follow
from this that revisions would have been a major source of policy error. Chart 3 shows the
credit-to-GDP gap using data available in real time and the latest data, suggesting that
revisions over this period were smaller for the gap, i.e. ratio relative to trend, than for the
ratio itself. This stems from the fact that revisions are autocorrelated, so they pull the trend
up and down as well as the ratio, reducing the effect of revisions on the gap estimate.
Chart 3: Initial and revised estimates of the
credit-to-GDP gap(a)(b)
Chart 4: Credit-to-GDP gaps calculated with
one- and two-sided HP filter(a)(b)
Source: ONS, Bank of England and Bank calculations.
(a) The credit-to-GDP gap is calculated as in footnote (a) in Chart 1. (b) Credit is defined here as debt claims on the UK private non-
financial sector. This includes all liabilities of the household sector
and private non-financial corporations’ loans and debt securities.
Source: ONS, Bank of England and Bank calculations.
(a) The credit-to-GDP gap is calculated as the percentage point difference between the credit to GDP ratio and its long-term trend,
where the trend is based on a one-sided/two-sided HP filter with a
smoothing parameter of 400,000. (b) Please see Chart 1 footnote (b) for the definition of credit.
Challenge ii): The choice of trend
Edge and Meisenzahl (2011) also argue that the ‘true’ underlying trend, measured using a
two-sided HP filter, may differ substantially from real-time estimates of the trend, measured
using a one-sided filter. The one-sided filter uses only data available up to each observation,
whereas the two-sided filter calculates the trend over the whole sample.11
They find that this
makes a substantial difference to estimates of the gap in the US and warn that this may lead
to policy errors.
Is this the case in the UK? Chart 4 shows that the credit gap calculated using a two-sided
filter is indeed very different from the one-sided one. But, as van Norden (2011) argues, it
does not follow that this will be a source of policy error. What matters is whether the real-
11 The two-sided filter chooses a trend that minimises a function of the deviations of the trend from the data series and the variation in the
trend growth rate for every point on the trend line. As more data points are added, previous estimates of the trend line will be updated to reflect the new data. In contrast, a one-sided filter produces a trend line which is purely backward looking.
-30
-20
-10
0
10
20
30
2004 2005 2006 2007 2008 2009 2010 2011 2012
Initial gap Latest gapPer cent of
GDP
-20
-15
-10
-5
0
5
10
15
20
25
30
1965 1971 1977 1983 1989 1995 2001 2007
One-sided filter
Two-sided filter Per cent of GDP
6
time credit gap estimates are useful for policymakers and our analysis above suggests that
they are. In fact, Drehmann et al. (2011)’s cross-country analysis shows that even if
policymakers did have the final credit gap estimates in real-time, it is not clear that they are a
better leading indicator of crises. Analysis in Borgy et al. (2009) suggests that the one-sided
filter leads the two-sided filter because, for the same smoothing parameter, it is influenced
more by the latest observation and hence more pro-cyclical. But since the trend lags the
actual observations, this implies that the credit gap crosses the one-sided trend earlier than the
two-sided trend, making the credit gap based on the one-sided trend more useful as a leading
indicator. Chart 4 shows that UK policymakers might indeed have found the real-time credit
gap more useful in the build-up of the late 1980s and mid-2000s credit boom as it signalled
the rapid credit expansion somewhat earlier.
We have also checked the robustness of our results with regard to the choice of trend.
Aikman et al. (2010) suggest that credit cycles are substantially longer than business cycles,
supporting the use of a smoothing parameter of 400,000 in the HP filtering. However, to
understand how the indicator varies with different assumptions about the credit cycle, we
calculated the trend with different smoothing parameters between 1,600 (usually assumed for
the business cycle) and 400,000. If the credit cycle is a medium-term phenomenon (i.e.
longer than 10 years in length), then using a parameter which assumes a much higher
frequency is unlikely to yield good results; nonetheless it is a useful robustness check. While
our analysis suggests that the power of the credit gap as an indicator is reduced when using
lower smoothing parameters, the results in broad terms still stand. The same applies to
different de-trending techniques, for example setting the trend equal to the line of best fit over
the preceding twenty years (Chart 5). This is in line with Borgy et al. (2009) whose results
on the whole do not differ materially between different methods to calculate the trend.
Chart 5: Credit-to-GDP gaps calculated with
different trends(a)(b)
Chart 6: Gap of broad and narrow total
private credit-to-GDP relative to long-term
trend (including intra-financial)(a)(b)(c)
Source: ONS, Bank of England and Bank calculations.
(a) The credit-to-GDP gap is calculated as the percentage point
difference between the credit to GDP ratio and its long-term trend, where the trend varies as above. (b) Please see Chart 1 footnote (b) for the definition of credit.
Source: ONS, Bank of England and Bank calculations.
(a) The credit-to-GDP gap is calculated as in footnote (a) in Chart 1. (b) Broad credit including intra-financial credit is defined here as debt claims on the UK private sector. It comprises broad credit (as
defined in Chart 1 footnote (b)) and loan, debt security and deposit
obligations of non-bank financial institutions (NBFIs). ONS data on NBFIs are not available before 1997. Before then, stable
-30
-20
-10
0
10
20
30
1965 1971 1977 1983 1989 1995 2001 2007
HP trend, lambda = 1600
HP trend, lambda = 400000
20 year moving average trend
Per cent of GDP
-40
-30
-20
-10
0
10
20
30
40
50
60
1965 1971 1977 1983 1989 1995 2001 2007
Broad private credit-to-GDP gap
Narrow private credit-to-GDP gap
Per cent of GDP
7
relationships between the ONS NBFI debt data and Bank of
England data on lending to other financial corporations are assumed.
(c) Narrow credit including intra-financial credit comprises narrow
credit as defined in Chart 1 footnote (c) and UK-resident monetary
financial institution sterling and foreign currency lending to other financial corporations.
Challenge iii): The definition of credit
We argued above that for the UK the credit gap based on both narrow and broad series of
credit to the private non-financial sector work well. The advantage of a broad series is that it
captures a wider range of sources of credit while the narrow series in the UK appears more
accurate and goes back further in time.
In addition to the private non-financial sector, intra-financial lending played an important role
ahead of the current crisis in the UK (Chart 6). A reason for excluding intra-financial system
exposures particularly from broad series is that lending from banks might go to households
via another intermediary and might therefore be counted twice if intra-financial credit was
included. However, Gai et al. (2011) show that greater complexity may contribute to time-
varying fragility in a financial network model. It may therefore be important to consider
intra-financial system lending in its own right or include it in the credit measures used to
calculate the credit gap.
3 Other indicators complementing the credit-to-GDP gap
The credit-to-GDP gap conveys an overall sense of the change in indebtedness of the
economy. But additional information including on the aggregate level of indebtedness and
indebtedness of different sectors is needed in order to form a view on how likely this is to
pose a threat to the financial system. This section discusses some areas where other
indicators may complement the credit-to-GDP gap, drawing on cross-country literature where
appropriate.
Complement i): What about levels?
The credit gap measure assumes that policy would want to be agnostic about the level of
credit in the economy. Recent research has, however, shown that the level may also matter.
With higher levels of leverage, the economy is more vulnerable to shocks and the
deflationary effect of subsequent deleveraging may be greater (Koo, 2008, Reinhart and
Rogoff, 2009). Furthermore Arcand et al. (2012) and Cecchetti and Kharroubi (2012) show
that there is an inverse U-shape relationship between financial system growth and economic
growth. This suggests that the absolute level of the credit-to-GDP ratio in itself would be of
interest to policymakers. In addition, it may be helpful to look at sectoral splits to understand
where exuberance might be building. For example, household debt-to-income and PNFC
debt-to-profits might be helpful in this regard. Chart 7 indicates that both would have
signalled increasing vulnerabilities in the UK prior to past crises. That said, UK households
have experienced little financial distress so far in this crisis given the exceptionally low level
of short-term interest rates and high incidence of floating rate mortgages (Bank of England
(2011a)).
8
Household debt-to-income ratios have also tended to increase prior to crises in other
advanced economies. For example, Mian and Sufi (2009) highlight the importance of
household debt in the US financial crisis, finding that US areas with a high share of subprime
borrowers experienced very rapid house price appreciation and growth in mortgage debt
before the crisis and very high default rates during the crisis. In subsequent (2011) work,
they have found that the decline in consumption was much stronger in counties with high
household leverage and conclude that the ‘household balance sheet channel was responsible
for a very large fraction of the decline in consumption during and after the recession’.
Chart 7: Household debt-to-income and
PNFC debt-to-profit ratios(a)
Chart 8: UK banks’ leverage(a)(b)
Source: ONS, Bank of England and Bank calculations.
(a) Gross debt as a percentage of a four-quarter moving sum of
disposable income and gross operating surplus of the respective
sector. ONS data on household and private non-financial corporate
(PNFC) debt are used from 1989 due to limited data availability.
Before then, a stable relationship is assumed between the ONS debt data and the Bank of England lending data. The household
disposable income and corporate gross operating surplus series are
adjusted for financial intermediation services indirectly measured (FISIM).
Source: Published accounts (a) Ratio of total assets to shareholders' claims.
(b) The data are a backwardly consistent sample of institutions
providing banking services in the United Kingdom in 2011. The sample includes the following financial groups: Barclays, HSBC,
LBG, National Australia Bank, Nationwide, RBS and Santander
UK. Where data are consistently available for the UK component of the banking group, these have been used. Northern Rock and
Bradford and Bingley were included in the chart up to 2007 and
2008 respectively.
Complement ii): The sources of credit matter
The credit-to-GDP gap is also silent on the sources of credit. However, the way lending is
funded is important. A highly leveraged financial system is likely to be more fragile than one
based on less leverage because there will be less equity to absorb losses materialising on
banks’ assets. At the same time, greater maturity transformation, for example when credit is
funded by high levels of short-term wholesale debt, is likely to make the financial system
more prone to liquidity crises. This highlights the importance of measures speaking to the
resilience of bank balance sheets.
For example, the leverage ratio of individual firms has been found to be a particularly useful
indicator in this regard prior to the current crisis and the system-wide leverage ratio picked up
sharply in the mid 2000s (Chart 8). Risk-based capital ratios also speak naturally to banking
system resilience but were less helpful in differentiating between banks which ended up
failing before the most recent crisis (see Haldane and Madouros (2012)). Moreover, the
0
100
200
300
400
500
600
0
40
80
120
160
200
1963 1970 1977 1984 1991 1998 2005 2012
Household debt, per cent of household income (right-hand scale)
PNFC debt, per cent of gross operating surplus (left-hand scale) Per centPer cent
0
10
20
30
40
50
60
70
60 66 72 78 84 90 96 02 08
Interquartile range
Range
Median
Ratio
9
leverage ratio has been found to be an informative early warning signal of crises in OECD
countries by Barrell et al. (2010).
Chart 9: Loan to deposit ratio(a)
Chart 10: Current account balance and long-
term real interest rate(a)(b)
Source: Bank of England and Bank calculations. (a) Sterling lending to deposit ratio is calculated as M4Lx divided
by retail deposits.
Sources: ONS, Bloomberg and Bank calculations. (a) As per cent of quarterly GDP. (b) 5 year real interest rates 5 years forward, derived from the Bank's index-linked government liabilities curve.
The flightiness of banks’ funding also matters. Shin and Shin (2011) and Hahm et al. (2012),
for example, emphasise the role of growing non-core liabilities as a source of financial
instability. A high and increasing loan-to-deposit ratio would then signal a weakening in
banks’ funding. Rising loan-to-deposit ratios were evident in many countries prior to this
crisis, including the UK (Chart 9), and the indicator also performed well in signalling
impending distress in some of the countries which suffered crises in East Asia in 1997-8.
Credit booms may then be funded by capital inflows from abroad. The resilience of the
domestic banking system would be undermined if it received foreign funding which turned
out to be flighty. If foreign lending was directly extended to end-users but domestic banks
took part in excess lending, resilience might also weaken. Large and persistent current
account deficits are therefore also often seen as a warning sign of building vulnerabilities (see
e.g. Barrell et al. (2010), Reinhart and Reinhart (2008)). Chart 10 shows that the UK ran a
current account deficit before the current and the small banks’ crisis. But countries with
current account surpluses are also not automatically isolated from banking crises because
their banks may hold risky claims on deficit countries.
Capital inflows might lead to low (real) long-term interest rates which in turn might fuel
excessive risk-taking via a search for yield. Borgy et al. (2009) have found that a decrease in
the long-term interest rate increases the probability of asset price booms. But real interest
rates may also be driven by other developments, making them harder to interpret. While the
real interest rate was low in the UK ahead of the current crisis, reflecting in part a global
saving glut that may have contributed to the crisis (see Bernanke (2005), Astley et al. (2009),
King (2009)), this was not the case prior to the small banks’ crisis (Chart 10).
0.0
0.5
1.0
1.5
2.0
2.5
3.0
1964 1970 1976 1982 1988 1994 2000 2006 2012
Sterling lending to deposit ratio Ratio
-6
-4
-2
0
2
4
6
-6
-4
-2
0
2
4
6
1960 1968 1976 1984 1992 2000 2008
Current account balance (left-hand scale) (a)
Five-year real interest rate five years forward (right-hand scale) (b)
Per cent of GDPPer cent
10
Chart 11: Real estate indicators(a)(b)
Chart 12: Residential mortgage loan-to-value
and loan-to-income ratios(a)
Sources: ONS, Nationwide, Halifax, Investment Property Databank
(IPD) and Bank calculations.
(a) The residential house price-to-rent index is the ratio between an
average of the Halifax and Nationwide house price indices and RPI
housing rent.
(b) The commercial property price-to-rent index is the ratio between
the IPD All Property Capital Growth Index and IPD All Property
Rental Value Index.
Sources: FSA Product Sales Database, CML and Bank calculations. (a) Excludes remortgagors.
Complement iii): The quality of credit matters
Besides the sources of credit, the uses of credit and the terms at which it is provided may also
matter. For example, credit growth may be less of a concern for financial stability if it was
used to finance investment projects that enhance the economy’s capacity to produce output
than if it was used to buy existing assets. House prices have typically been linked to financial
crises (see, among others, Barrell et al. (2010), Drehmann et al. (2010), Claessens et al.
(2011), Mendoza and Terrones (2008) and Riiser (2005)) and they tend to lead volume-based
credit indicators. In the UK, peaks of the deviation of house prices relative to their long-term
trend as well as relative to incomes and rent have tended to lead episodes of banking system
distress over past decades (Chart 11). Loan–to-income multiples in new mortgages also
picked up ahead of the current crisis. But while loan-to-value ratios were a good leading
indicator of stress in the United States ahead of the current crisis, they remained below their
long-run average before 2007 in the UK (Chart 12). High loan-to-value mortgages are likely
to be more risky in the US than in the UK because several states in the US protect borrowers
from personal liability if a property is foreclosed and sold at less than the mortgage value.
Equity prices, on the other hand, tend to do less well in signalling financial stress as
suggested in our analysis for the UK and as documented in Drehmann et al. (2011). While
Kaminsky and Reinhart (1999) and Borio and Lowe (2002) find some evidence that the stock
market can be a useful indicator for banking crises, there are also shown to be more false
signals than from other indicators. Not all equity price bubbles lead to widespread financial
stress (e.g. dotcom crash) and equity markets may be sensitive to business cycle movements.
0
20
40
60
80
100
120
140
160
180
1987 1991 1995 1999 2003 2007 2011
House price-to-rent ratio
Commercial property price-to-rent ratio
Index, 1987-2006average = 100
0
20
40
60
80
100
0.0
1.0
2.0
3.0
4.0
5.0
1979 1985 1991 1997 2003 2009
Median LTI (right-hand scale)
Median LTV (left-hand scale)Per cent Multiple
11
But if credit was used to speculate in rising equity markets as for example in the Great
Depression, this would clearly be a cause for concern.
Lending spreads (Chart 13) may help to differentiate between demand and supply conditions
in credit markets: before a crisis, we would expect spreads to be low if risk premia were
unsustainably compressed due to exuberant credit supply (e.g. before the current crisis),
whereas they may be at normal levels if credit growth reflected a greater balance between
demand and supply factors (e.g. before the secondary banking crisis and the small banks’
crisis). An advantage is that they are available in a timely manner. However, judging what is
meant by ‘unsustainably low’ spreads is not straightforward.
Chart 13: Spreads on new lending(a)(b)
Chart 14: Real and nominal credit growth(a)
Source: Bank of England, Council of Mortgage Lenders, British
Bankers' Association, Bank of America Merrill Lynch, de Montfort
University, Bloomberg and Bank calculations.
(a) The mortgage spread is a weighted average of quoted mortgage
rates over safe rates, using 90% LTV 2-year fixed rate mortgages
and 70% LTV tracker, two- and five-year fixed rate mortgages.
Spreads are taken relative to gilt yields of matching maturity for
fixed rate products until August 2009, after which spreads are taken
to OIS of matching maturity. Spreads are taken to Bank Rate for the
tracker product.
(b) The corporate lending spread is a weighted average of UK
investment grade company bond spreads to gilts (adjusted for any
embedded option features such as convertibility into equity), SME
lending rates to Bank Rate and CRE lending rates to Bank Rate.
Source: ONS, Bank of England, Thomson-Reuters Datastream and
Bank calculations.
(a) Twelve-month growth rate of nominal credit.
(b) Credit is defined here as debt claims on the UK private non-
financial sector. This includes all liabilities of the household and
not-for-profit sector and private non-financial corporations’ loans
and debt securities excluding derivatives, direct investment loans and
loans secured on dwellings.
(c) Real credit growth is deflated using the GDP deflator.
Complement iv): The release phase
BCBS (2010) suggests that the countercyclical capital buffer should be released when the
credit cycle turns. Chart 1 suggests that besides being a reasonable indicator for exuberance
in credit growth, the credit-to-GDP gap would also have indicated the turning of the credit
cycle in the UK. However, it may not be a timely indicator for policy easing in all
circumstances. For example, Drehmann et al. (2011) show that the credit-to-GDP gap is
slow to decline once crises materialise. A potential reason for this finding is that the stock of
credit may not fall immediately in a downturn because corporates may have undrawn credit
lines available. Also GDP growth may fall, potentially even leading to an increase in the
ratio. The growth rate of credit variables may provide a more timely alternative to the credit
0
50
100
150
200
250
300
350
400
1997 2000 2003 2006 2009 2012
Blended UK mortgage spread
Blended UK corporate lending spread
Basis points
-30
-20
-10
0
10
20
30
40
1964 1970 1976 1982 1988 1994 2000 2006 2012
Nominal credit growth (b)
Real credit growth (c)
Percentage points
12
gap in identifying turning points of the financial cycle. Evidence presented in Drehmann et
al. (2011) suggests that credit growth tends to reverse prior to the onset of stress – a finding
which is in line with evidence from past episodes of banking system distress in the UK (Chart
14). In addition, an increase in lending spreads might indicate a tightening of credit
conditions and might be a useful signal to ease policy.
Bank of England (2013) argues, however, that a downturn in the credit cycle alone is not
sufficient for easing policy. In addition, information on banks’ resilience in the form of
capital adequacy and market-based indicators is required. While the former give the
policymaker an understanding of banks’ ability to absorb prospective losses from any
remaining threats and support lending during a downturn, the latter give an indication on
whether markets would allow banks to reduce their capital ratios to increase lending.
Measures of banks’ profitability may also be informative because profits allow banks to build
capital and are their first line of defence to absorb losses.
If banks’ balance sheets appear resilient, banks’ debt spreads remain low and banks’ equity
prices do not fall markedly, policy easing might be appropriate. However, when market-
based indicators such as banks’ debt spreads increase sharply signalling concerns over banks’
solvency, this might not be the time to release capital buffers. Indeed, in such circumstances
it may be more appropriate for banks to increase levels of capital which may lead to a
reduction in funding costs (see Bank of England (2013)). For the UK, CDS premia and
interbank spreads rose sharply at the beginning of the current crisis (Chart 15). Gropp et al.
(2002) and Drehmann et al. (2011) also find that bank credit spreads tend to coincide with
stress events. This conditionality on stress materialising or not makes it difficult to test
indicators for release in practice, particularly on time series when the macroprudential policy
regime was not in operation.
Chart 15: Bank funding spreads(a)
Chart 16: ROC curve for useful and useless
indicators
Sources: Markit Group Limited, published accounts and Bank
calculations.
(a) Average of major UK banks' five-year senior CDS premia,
weighted by end-year total assets. Includes Nationwide from July
0
50
100
150
200
250
300
350
0
100
200
300
400
500
600
700
1972 1980 1988 1996 2004 2012
Three-month UK interbank spread (right-hand scale)
Senior CDS premia (left-hand scale)
Basispoints
Basispoints
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Sign
al r
atio
Noise ratio
ROC curve for useful indicator
ROC curve for useless indicator
Noise ratio at signal-maxmising threshold
Signal ratio at noise-minimisingthreshold
AUROC (yellow area)
13
2003.
4 Comparison of different indicators
In this section, we conduct a simple analysis of the relative usefulness of some of these
indicators in the UK context. We limit the analysis to those indicators for which we have
data covering at least the three past episodes of banking system distress. For each indicator,
we use quarterly data from 1969 Q1 until 2009 Q112
. In the analysis, we take a non-
parametric approach, evaluating indicators by their ability to provide good signals of
forthcoming crises and to avoid false alarms. There are various different ways to use this
information to rank indicators. As discussed below, we choose three criteria: the area under
the receiver operator characteristic curve (AUROC), the signal ratio at the threshold which
minimises the noise ratio and the noise ratio at the threshold that maximises the signal ratio.
These rankings should be treated with appropriate caution given the low number of episodes
of banking system distress in our sample. In future, we seek to extend this methodology to a
larger set of countries, notwithstanding the potential drawbacks of this approach highlighted
above.
A perfect early warning indicator of financial crises from a policymaking perspective would
emit a signal ahead of every crisis and at no other time. To be useful to a policymaker, it
must emit this signal sufficiently far in advance of a crisis for mitigating actions to be
implemented and to have an effect.13
With this in mind, we have categorised the
observations for each indicator thus:
A: If an indicator is above a threshold value and a crisis occurs at any time between
one year afterwards and four years afterwards, the observation is categorised as a
good signal. As a result, there are twelve potential good signals for each banking
crisis, giving 36 possible signals overall, rather than just one for each of the three
crises (although these are, of course, not independent observations). This is meant to
capture the intuition that a persistent signal may be more valuable to a policymaker
than a signal that just flashes once before a crisis. That said, there may be
12 The leverage series we have is annual so we interpolated it to obtain quarterly data. 13 BCBS (2010) suggests that policymakers preannounce an increase in the countercyclical capital buffer by twelve months. Of course, one hopes that policy will be able to prevent risk building to the levels at which crises are likely.
14
diminishing marginal returns to signals ahead of any one crisis. To capture this, we
give more weight to the first four quarters for which a signal is emitted than to any
following observations ahead of a period of stress (the first is weighted by a five, the
second by a four,...the fifth and any others by a one). This weighting is of course
arbitrary and could be developed in future work.14
B: If an indicator is below a threshold value and a crisis occurs at any time between
one year afterwards and four years afterwards, the observation is categorised as a
missed signal (type 1 error).
C: If an indicator is above a threshold value and a crisis does not occur at any time
between one year afterwards and four years afterwards, the observation is categorised
as a false alarm (type 2 error).
D: If an indicator is below a threshold value and a crisis does not occur at any time
between one year afterwards and four years afterwards, the observation is categorised
as a good silence.
Note that as the threshold chosen changes, the values for A to D will change (a higher
threshold will be associated with higher values for B and D, missed signals and good
silences, and lower values for A and C, good signals and false alarms).
Two more methodological points are worth highlighting. First, we have chosen each
indicator in such a way that theory would lead us to expect a high value of the indicator to
presage a crisis and a low value the absence of crises. This meant reversing the sign of some
variables (e.g. we used the current account deficit, not the current account balance). Second,
to prevent an observation from being called a good signal or false alarm during or in the year
preceding a crisis, we exclude these observations.
As the numbers of crisis and non-crisis observations are fixed, we can condense these
summary statistics further into the signal ratio,
, and the noise ratio,
,
without losing any relevant information. Note that with 36 potential crisis signals (twelve of
which have a weight greater than one), SR and NR can take a wide range of values rather
than just 0, ⅓, ⅔ and 1, as would be the case if indicators could only have a single binary
signal in relation to each of the three crisis episodes. For any given threshold, the
policymaker would prefer an indicator with a high signal ratio and a low noise ratio.
However, there will be a trade-off between these two desirable features. For low thresholds,
both the signal ratio and the noise ratio are likely to be high as the indicator emits signals
most of the time. The opposite scenario applies for high threshold values. If costs of
macroprudential interventions are low and benefits high, policymakers may prefer a low
threshold value to avoid type 1 errors rather than type 2 errors.
14 The weighting scheme does not have marked impacts on the AUROC, signal ratio and noise ratios shown in Table 1 below. The
AUROCs and signal ratios tend to be somewhat higher, and noise ratios somewhat lower, with the weighting scheme. It has little bearing on the relative performance of the indicators against these metrics.
15
The receiver operating characteristic (ROC) curve summarises this trade-off. It plots the
noise ratio against the signal ratio. As the threshold value falls, both the noise and the signal
ratio rise, so the ROC curve slopes upwards. The ROC curve associated with a useless
indicator (i.e. an indicator for which the probability of a crisis if the indicator exceeds a
threshold is equal to the unconditional probability of a crisis) would be a 45 degree line, as
shown in the chart15
. A perfect indicator would give no signals above a high threshold (i.e.
zero signal and noise ratios), only signals below a low threshold (i.e. signal and noise ratios
equal to one) and neither type I nor type II errors in between (i.e. a signal ratio equal to one
and noise ratio equal to zero). In other words, it would trace out the left hand and top edges
of the chart above. Indicators which are useful to policymakers lie significantly above the 45
degree line (e.g. the red line in the chart).
Our first criterion for ranking the indicators, the area under the ROC curve (AUROC), does
not rely on any particular choice of threshold. Rather, it summarises the degree to which the
signal ratio exceeds the noise ratio for all thresholds. An indicator with an AUROC equal to
one would be a perfect early warning indicator (the whole of the chart above would be under
the ROC curve), while a useless indicator would have an AUROC equal to a half (half the
chart above is under the 45 degree line; see Chart 16). In general, the higher is an indicator’s
AUROC, the more useful it will be for a policymaker.
In our calculations, we have approximated the AUROC by: (i) plotting the noise and signal
ratios for the maximum and minimum observed value of the indicator in question and
nineteen points in between; (ii) calculating the trapezoidal area under the curve for each of
the twenty resulting intervals (equal to the product of the change in the noise ratio and the
average signal ratio); and (iii) aggregating. The AUROC is shown as the shaded area in
Chart 16.
The second and third criteria – the signal ratio at the noise-minimising threshold and the
noise ratio at the signal-maximising threshold – are also marked on Chart 16. The second
criterion would be chosen by a policymaker who seeks to minimise type 2 errors, while the
third criterion would be chosen by a policymaker who seeks to minimise type 1 errors. In
future work, we plan to model policymaker preferences over type 1 and type 2 errors. This
would allow us to choose optimal thresholds for each indicator and then rank the indicators
by their ability to satisfy these preferences.
As well as calculating these three statistics, we also derive their statistical significance using
the recursive bootstrap method (MacKinnon, 2006, is a very helpful guide). Each indicator is
modelled as an AR(ρ) process where ρ is chosen using the Schwartz Information Criterion.
The residuals from each of these regressions are then rescaled by a hat matrix to ensure that
their variance is unbiased. These are sampled randomly and, together with the coefficients
15
A useless indicator has the following properties: ( ) ( ) and ( ) ( ),
where is the indicator value, is the threshold and is a crisis. For a useless indicator, (( ) )
( )
( ) ( )
( ) ( ) and
(( ) )
( )
( ) ( )
( ) ( ), so the signal and
noise ratios are equal.
16
from the AR(ρ) regressions, are used to construct bootstrap samples. The actual AUROCs,
signal ratios at the noise-minimising threshold and noise ratios at the signal-maximising
threshold are then compared to the same statistics for the bootstrapped series. The ranking of
the actual statistics in the distributions for the simulated data gives the p value.
This method works for autoregressions for which the residuals are not heteroskedastic. For
some indicators, this is not the case. In these cases, we use the recursive wild bootstrap. This
is the same method, except that the residuals are kept in the same order but multiplied by
random draws from the Rademacher distribution (1 with p=0.5, -1 with p=0.5).
Table 1 shows the results. The credit gap indicators tend to perform reasonably well, with
high and statistically significant AUROCs. The signal ratios at the noise-minimising
threshold are all statistically significant and comfortably above zero, while the noise ratios at
the signal-maximising threshold are reasonably low and statistically significant. The precise
sectors covered do not make a lot of difference in this sample. Nor does the choice between
broad and narrow credit. This is not surprising, as they are highly correlated. The flow-
based credit measures tend to perform less well than the gap metrics, with real growth rates
of credit outperforming nominal growth rates, though not by much.
Of the other indicators, the real house price gap performs particularly well, with the highest
AUROC of the whole sample. That said, it might not be very useful for a policymaker who
cares a lot about type II errors (because the signal ratio at the noise-minimising threshold is
low and insignificant). The equity price gap and leverage ratio do not perform well.
Table 1: Comparing indicators through AUROC, minimum noise and maximum signal
ratios(a)
Source: ONS, Bank of England, Thomson-Reuters Datastream, Investment Property Databank, Global Financial Database, Halifax,
Ranking method AUROC
Indicator Threshold Signal ratio Threshold Noise ratio
AGGREGATE GAPS
Broad HH and PNFC credit gap 0.87* 12.5 0.39** -2.7 0.48**
Narrow HH and PNFC credit gap 0.84* 9.4 0.33** -1.4 0.44**
Broad HH, PNFC and OFC credit gap 0.79 22.9 0.41** -2.3 0.79*
Narrow HH, PNFC and OFC credit gap 0.87** 13.6 0.45** -2.3 0.51**
AGGREGATE GROWTH RATES
Nominal broad HH and PNFC credit growth 0.69 26.4 0.08 7.9 0.84
Nominal narrow HH and PNFC credit growth 0.71 24.2 0.08 8.6 0.73*
Nominal broad HH, PNFC and OFC credit growth 0.74 24.8 0.14 8.0 0.88
Nominal narrow HH, PNFC and OFC credit growth 0.73 25.5 0.14 8.9 0.69**
Real broad HH and PNFC credit growth 0.77 19.8 0.08 -1.6 0.95
Real narrow HH and PNFC credit growth 0.81** 17.8 0.21** -0.4 0.90
Real broad HH, PNFC and OFC credit growth 0.82** 17.2 0.35** -1.0 0.95
Real narrow HH, PNFC and OFC credit growth 0.79* 19.9 0.14 -0.4 0.93
OTHER INDICATORS
HH DTI gap 0.85* 15.7 0.50** -1.7 0.63
PNFC DTP gap 0.82* 68.6 0.00 -20.0 0.48**
OFC credit-to-GDP gap 0.60 23.5 0.21 -0.4 1.00
Current account deficit 0.67 3.9 0.18* -3.0 0.99
Loan-to-deposit ratio gap 0.78 0.1 0.32** 0.0 0.85
Leverage ratio 0.48 26.4 0.30** 12.2 1.00
Real house price gap 0.88** 33.7 0.21 -3.5 0.58**
Real commercial property price gap 0.83* 15.0 0.53*** -4.3 0.80
Real equity price gap 0.32 110.7 0.00 -34.8 0.98
Corporate bond spread 0.61 3.2 0.00 0.0 1.00
Minimum noise ratio Maximum signal ratio
17
Nationwide, published accounts and Bank calculations.
(a) *, ** and *** denote statistical significance at the 10%, 5% and 1% level respectively.
5 Using the indicators in practice: a case study
We have argued above that more than one indicator is required to set the countercyclical
capital buffer. This begs the question of how the dominant signal that should guide policy
might be extracted from the set of indicators. There are different ways to combine
information from different indicators, including statistical approaches that allow for
weighting different indicators, ‘heatmaps’ that rely on thresholds and pure judgement.
Regarding the former, principal component analysis or logit regressions may for example be
used to combine indicators. Heatmaps that assign a colour to each indicator depending on the
indicator’s signal as defined by statistical rules may be a useful visualisation device regarding
the overall message from the set of indicators. However, an element of judgement is still
required in any approach as the importance of different indicators may vary over time and
further information, such as market or supervisory intelligence, would also be important. A
set of indicators should be seen as the starting point for analysis. If indicators in this set flag
an issue, this should be investigated further.
Both Switzerland and Norway have opted for a ‘guided discretion’ approach with neither
relying on a purely mechanical relationship between indicators and setting the CCB.16
In a
similar spirit, the UK’s interim FPC notes in Bank of England (2013) that “the greater the
degree of imbalance as measured by the core indicators, the more homogeneous the picture
that the different indicators convey, and the more consistent that picture is with market and
supervisory intelligence, the more likely it is that the FPC will adjust the CCB and SCRs in
response”.
The remainder of this section discusses the signals given in late 2003 by some of the
indicators discussed in this paper. We choose 2003 in order to understand whether the
indicators were useful in signalling unsustainable credit growth and threats to resilience well
ahead of the start of the current crisis. Overall, the indicators gave a mixed signal.
The Basel III credit-to-GDP gap stood at 9.1% in Q4 2003, which under the Basel III
guidance would have translated into a buffer of almost 2.5%. Including intra-financial credit
doubles the gap. The Basel III gap had grown rapidly in the preceding years, reaching a peak
in Q3 2003, but declined in the last quarter of 2003. While the rapid increase in the early
2000s provided a clear signal of exuberant credit conditions, the decline in Q4 2003 might
have given pause for thought. Nominal credit was growing at above 10% on the previous
year in late 2003 and had been growing at this pace for a number of years. While high, this
was not particularly high compared with the growth observed before the small banks’ crisis in
the early 1990s. This is, however, where the level of the series also matters: the growth rates
observed might not have given a strong signal of being unsustainable in any particular quarter
given historical evidence, but the indebtedness of the economy was growing given persistent
growth. In terms of ratios, household debt to income was at an all-time high and PNFC debt-
16
See Swiss National Bank (2013) and Norges Bank (2013).
18
to-profit close to its peak, although it decreased somewhat through 2003. This pattern was
reflected in property prices as well: while residential house prices were booming, the
commercial real estate market had cooled somewhat.
This rise in the economy’s indebtedness started to show in banks’ balance sheets: banks’
leverage picked up in 2003, but remained within the bounds of historical observations and
banks increased reliance on non-core funds with the loan-to-deposit ratio back at the highs
seen prior to the small banks’ crisis. Banks’ profitability strengthened as lending spreads
ticked up and banks’ CDS premia fell. The change in these indicators may have signalled a
strengthening in banks’ resilience at the time rather than an emerging underpricing of risk:
banks’ profits had taken a dip and CDS premia had increased the year before.
The current account deficit narrowed in late 2003 but the deficit had persisted for a number of
years, contributing to growing stock imbalances. The real long-term interest rate was stable
around 2.25%, up from lows observed in the late 1990s and 2000-1 and hence by itself not
suggestive of a search for yield or the need for policy tightening.
While credit conditions could, on balance, have been judged to be buoyant in late 2003 and in
preceding quarters, and there was some evidence that banks’ funding structures were
becoming less stable, the indicators did not suggest significant underpricing of risk in a
historic context. These are exactly the conditions under which market and supervisory
intelligence and policymaker judgement are critical: Bank of England (2003) notes that
market contacts describe some investors as willing to take more credit, interest rate or
exchange rate risk in order to increase returns in the short run, i.e. an early indication of
search for yield behaviour. By 2005, the indicators themselves were painting a more
homogenous picture and would have given a clear signal of unsustainable credit conditions
and material threats to banks’ resilience.
6 Conclusion
This paper reviews the performance in the UK of the credit-to-GDP gap proposed under
Basel III for the countercyclical capital buffer. We find that irrespective of criticisms, it
provided timely signals for policy tightening over the past 50 years in the UK.
We also discuss how to complement the credit gap in the UK context. In particular, we find
that flow- and market-based indicators may complement the credit gap for decisions to
release the countercyclical capital buffer. Other indicators, such as property prices and
sectoral credit ratios, may give an indication of what projects the lending is used to finance
and hence of its quality. Indicators related to property prices were particularly important
during past credit cycles in the UK. In addition, indicators related to banks’ balance sheets,
e.g. leverage ratio or loan-to-deposit ratios, would be needed to understand how a credit
boom is funded.
But further work in this area is required: we do not understand well how indicators will
change as policy is implemented, i.e. to what extent Goodhart’s law is likely to apply.
19
Moreover, our knowledge on the counterfactual path of the economy if the signals from the
indicators are heeded is limited. And in the future, if macroprudential policy is successful in
reducing the frequency and severity of financial crises, assessing the usefulness of indicators
cannot rely on whether the indicators forecast crises. Finally, our analysis focused on
indicators for setting capital tools. Indicators for other macroprudential tools, such as
liquidity or structural tools, would also need to be developed.
References
Aikman, D, Haldane, A G and B D Nelson (2010), ‘Curbing the credit cycle’, prepared for
the Columbia University Centre on Capital and Society Annual Conference, New York,
November 2010.
Alessi, L and Detken, C, ‘Quasi real time early warning indicators for costly asset price
boom/bust cycles: A role for global liquidity’, European Journal of Political Economy 27.3
(2011): 520-533.
Arcand, Jean Louis, Enrico Berkes, and Ugo Panizza (2012), ‘Too much finance’, IMF
Working Paper WP/12/161.
Astley, M, Giese, J, Hume, M and Kubelec, C (2009), ‘Global imbalances and the financial
crisis’, Bank of England Quarterly Bulletin, 2009 Q3.
Bank of England (2003), Financial Stability Review, December 2003.
Bank of England (2009), ‘The role of macroprudential policy’, Discussion Paper, November
2009.
Bank of England (2011a), Financial Stability Report, June 2011.
Bank of England (2011b), ‘Instruments of macroprudential policy: a discussion paper’,
December
Bank of England (2013), ‘The Financial Policy Committee’s powers to supplement capital
requirements: A Draft Policy Statement’, January.
Barrell, R., Davis, EP., Karim, D. and Liadze, I. (2010), ‘Bank regulation, property prices and
early warning systems for banking crises in OECD countries’, Journal of Banking & Finance
34 (9): 2255- 2264
BCBS (2010), ‘Countercyclical capital buffer proposal – consultative document’, July 2010.
Bernanke, B (2005), ‘The global saving glut and the US current account deficit’, speech on
10 March.
Borgy, V, Clerc, V, and J-P Renne (2009), ‘Asset-price boom-bust cycles and credit: what is
the scope of macroprudential regulation?’, Banque de France Working Paper No 263.
20
Borio, C (2011), ‘Implementing a macroprudential framework: blending boldness and
realism’, Capitalism and Society, Vol. 6, Issue 1, Article 1.
Borio, C, and Lowe, P (2002), ‘Asset prices, financial and monetary stability: exploring the
nexus’, BIS Working Papers No. 114.
Borio, C, and Lowe, P (2004), ‘Securing sustainable price stability: should credit come back
from the wilderness? ’, BIS Working Papers No. 157.
Cecchetti, S and Kharroubi, E (2012), ‘Reassessing the impact of finance on growth’, BIS
Working Paper No. 381.
Claessens, S, Kose, MA, and Terrones, ME (2011), ‘How do business and financial cycles
interact?’, IMF Working Paper 11/88.
Committee on the Global Financial System (2010), ‘Macroprudential instruments and
frameworks: a stocktaking of issues and experiences’, CGFS Publications no. 38.
Dell'Ariccia, G, Igan, D, Laeven, L, Tong, H, Bakker, B, Vandenbussche, J (2012), ‘Policies
for Macrofinancial Stability: How to Deal with Credit Booms’, IMF Staff Discussion Notes
No. 12/06
Drehmann, M, Borio, C, Gambacorta, L, Jimenez, G and Trucharte, C (2010),
‘Countercyclical capital buffers: exploring options’, BIS Working Paper 317.
Drehmann, M, Borio, C and Tsatsaronis, K (2011), ‘Anchoring countercyclical capital
buffers: the role of credit aggregates’, BIS Working Paper No 355.
Edge, R and Meisenzahl, R (2011), ‘The Unreliability of Credit-to-GDP Ratio Gaps in Real
Time: Implications for Countercyclical Capital Buffers’, International Journal of Central
Banking.
Financial Stability Board, International Monetary Fund, Bank for International Settlements
(2011), ‘Macroprudential policy tools and frameworks: progress report to G20’, 27 October.
Gai, P, Haldane, A and Kapadia, S (2011), ‘Complexity, concentration and contagion’,
Journal of Monetary Economics, 58 (5): 453-470.
Gropp, R, Vesala, J, and Vulpes, G (2002), ‘Equity and bond market signals as leading
indicators of bank fragility’, ECB Working Paper Series, No. 150.
Hahm, J-H, Shin, H S and Shin, K (2012), ‘Non-core bank liabilities and financial
vulnerabilities’, mimeo.
Haldane, AG, and Madouros, V (2012), ‘The dog and the frisbee’, Speech given at the
Federal Reserve Bank of Kansas City’s 36th
economic policy symposium.
21
International Monetary Fund (2011), ‘Macroprudential policy: an organising framework’,
March.
Kaminsky, GL and Reinhart, CM (1999), ‘The Twin Crises: The Causes of Banking and
Balance-of-Payments Problems’, American Economic Review 89(3): 473–500.
Kindleberger, CP (1978), Manias, Panics and Crashes: A History of Financial Crises, 1st
edition, New York: Basic Books, 1978.
King, M A (2009), The Governor’s speech to the CBI Dinner, Nottingham, January,
Koo, R (2008), The Holy Grail of Macroeconomics: Lessons from Japan’s Great Recession,
Wiley.
Logan, A (2000), ‘The early 1990s small banks’ crisis: leading indicators’, Bank of England
Financial Stability Review, December, pages 130–45.
MacKinnon, J (2006), ‘Bootstrap Methods in Econometrics’, The Economic Record, vol 82,
special issue, September.
Mendoza, E and Terrones, M (2008), ‘An anatomy of credit booms: Evidence from macro
aggregates and micro data’, NBER Working Paper no. 14049.
Mian, A and Sufi, A (2009), ‘The consequences of mortgage credit expansion: evidence from
the US mortgage default crisis’, Quarterly Journal of Economics, no 124, pp 1449–96.
Mian, A and Sufi, A (2011), ‘House prices, home equity-based borrowing and the US
household leverage crisis’, American Economic Review, no 101, pp 2132–56.
Minsky, HP (1972), “Financial Instability Revisited: The Economics of Disaster.”
Reappraisal of the Federal Reserve Discount Mechanism 3 (1972): 97-136.
Norges Bank (2013), ‘Criteria for an appropriate countercyclical capital buffer’, Norges Bank
Papers No. 1.
Orphanides, A and van Norden, S (2002), ‘The unreliability of output-gap estimates in real
time’, Review of Economics and Statistics 84.4: 569-583.
Reid, M (1982), The secondary banking crisis, 1973–1975, Macmillan Press.
Reinhart, C and Reinhart, V (2008), ‘Capital flow bonanzas: an encompassing view of the
past and present’, NBER Working Paper no. 14321.
Reinhart C and Rogoff K (2009), This time is different: Eight Centuries of Financial Folly,
Princeton University Press.
22
Riiser, M (2005), ‘House prices, equity prices, investment and credit – what do they tell us
about banking crises? A historical analysis based on Norwegian data’, Norges Bank
Economic Bulletin 05 Q3, pp. 145-154
Schularick, M and Taylor A (2012), ‘Credit booms gone bust: Monetary policy, leverage
cycles, and financial crises, 1870–2008”, American Economic Review, no 102, pp 1029–61.
Shin, HS and Shin, K (2011): “Procyclicality and monetary aggregate”. NBER Working
Paper No. 16836.
Swiss National Bank (2013), ‘Implementing the countercyclical capital buffer in Switzerland:
concretising the Swiss National Bank’s role’.
Van Norden, S (2011): ‘Discussion of “The Unreliability of Credit-to-GDP Ratio Gaps in
Real Time: Implications for Countercyclical Capital Buffers”, International Journal of
Central Banking.
23
Broad HH and PNFC credit gap Narrow HH and PNFC credit gap
Source: ONS, Bank of England and Bank calculations.
Source: Bank of England and Bank calculations.
Broad HH, PNFC and OFC credit gap Narrow HH, PNFC and OFC credit gap
Source: ONS, Bank of England and Bank calculations.
Source: Bank of England and Bank calculations.
Nominal broad HH and PNFC credit
growth
Nominal narrow HH and PNFC credit
growth
Source: ONS, Bank of England and Bank calculations.
Source: Bank of England and Bank calculations.
-20
-15
-10
-5
0
5
10
15
20
25
30
1965 1973 1981 1989 1997 2005
Broad HH and PNFC credit gap Per cent of GDP
-20
-15
-10
-5
0
5
10
15
20
25
30
1965 1973 1981 1989 1997 2005
Narrow HH and PNFC credit gap Per cent of GDP
-40
-30
-20
-10
0
10
20
30
40
50
60
1965 1973 1981 1989 1997 2005
Broad HH, PNFC and OFC credit gap Per cent of GDP
-40
-30
-20
-10
0
10
20
30
40
50
60
1965 1973 1981 1989 1997 2005
Narrow HH, PNFC and OFC credit … Per cent of GDP
-10
-5
0
5
10
15
20
25
30
35
40
1964 1972 1980 1988 1996 2004 2012
Nominal broad HH and PNFC credit growth
Per cent of GDP
-10
-5
0
5
10
15
20
25
30
1964 1972 1980 1988 1996 2004 2012
Nominal narrow HH and PNFC credit growth
Per cent of GDP
24
Nominal broad HH, PNFC and OFC
credit growth
Nominal narrow HH, PNFC and OFC
credit growth
Source: ONS, Bank of England and Bank calculations.
Source: Bank of England and Bank calculations.
Real broad HH and PNFC credit growth Real narrow HH and PNFC credit growth
Source: ONS, Bank of England and Bank calculations.
Source: ONS, Bank of England, Thomson-Reuters Datastream and
Bank calculations.
Real broad HH, PNFC and OFC credit
growth
Real narrow HH, PNFC and OFC credit
growth
-10
-5
0
5
10
15
20
25
30
35
40
1964 1972 1980 1988 1996 2004 2012
Nominal broad HH, PNFC and OFC credit growth
Per cent of GDP
-10
-5
0
5
10
15
20
25
30
35
40
1964 1972 1980 1988 1996 2004 2012
Nominal narrow HH, PNFC and OFC credit growth
Per cent of GDP
-30
-20
-10
0
10
20
30
1964 1972 1980 1988 1996 2004 2012
Real broad HH and PNFC credit growth
Per cent of GDP
-30
-20
-10
0
10
20
30
1964 1972 1980 1988 1996 2004 2012
Real narrow HH and PNFC credit growth
Per cent of GDP
-30
-20
-10
0
10
20
30
1964 1972 1980 1988 1996 2004 2012
Real broad HH, PNFC and OFC credit growth
Per cent of GDP
-30
-20
-10
0
10
20
30
1964 1972 1980 1988 1996 2004 2012
Real narrow HH, PNFC and OFC credit growth
Per cent of GDP
25
Source: ONS, Bank of England, Thomson-Reuters Datastream and
Bank calculations.
Source: Bank of England, Thomson-Reuters Datastream and Bank
calculations.
HH debt-to-income gap PNFC debt-to-profit gap
Source: ONS, Bank of England and Bank calculations.
Source: ONS, Bank of England and Bank calculations.
OFC credit-to-GDP gap Current account deficit
Source: ONS, Bank of England and Bank calculations.
Source: ONS and Bank calculations.
Loan to deposit ratio gap Leverage ratio
Source: Bank of England and Bank calculations.
Source: Published accounts and Bank calculations.
-30
-20
-10
0
10
20
30
1965 1973 1981 1989 1997 2005
HH DTI gapPer cent
-100
-80
-60
-40
-20
0
20
40
60
80
100
1965 1973 1981 1989 1997 2005
PNFC DTP gap Per cent
-20
-10
0
10
20
30
40
1965 1973 1981 1989 1997 2005
OFC credit-to-GDP gapPer cent of
GDP
-6
-4
-2
0
2
4
6
1963 1971 1979 1987 1995 2003 2011
Current account deficit Per cent
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
1965 1973 1981 1989 1997 2005
Loan to deposit ratio gap Ratio
0
5
10
15
20
25
30
35
40
45
50
1965 1973 1981 1989 1997 2005
Leverage ratio Ratio
26
Real house price gap Real commercial property price gap
Source: Nationwide, Halifax and Bank calculations.
Source: Investment Property Databank and Bank calculations.
Real equity price gap Corporate bond spread
Source: Thomson-Reuters Datastream and Bank calculations.
Source: Global Financial Database and Bank calculations.
-30
-20
-10
0
10
20
30
40
50
1969 1977 1985 1993 2001 2009
Real house price gapPer cent of
GDP
-40
-30
-20
-10
0
10
20
30
40
1969 1977 1985 1993 2001 2009
Real commercial property price gap
Per cent of GDP
-60
-40
-20
0
20
40
60
80
100
120
1969 1977 1985 1993 2001 2009
Real equity price gapPer cent of
GDP
0
1
2
3
4
5
6
7
1965 1973 1981 1989 1997 2005
Corporate bond spreadPer cent